Bloomberg reported on the 25th that Meta, the parent company of Facebook and Instagram, is in talks to adopt Google artificial intelligence (AI) chips.
According to Bloomberg, Meta, which has been using Nvidia graphics processing units (GPUs), is reviewing a plan to build data centers by adopting Google's in-house-developed tensor processing units (TPUs) on a multibillion-dollar scale in 2027. It also said Meta could rent and use TPUs from Google Cloud next year.
Bloomberg assessed that if the transaction is finalized, Google TPUs would help establish themselves as an alternative to Nvidia chips. Earlier, Google signed a contract to supply up to 1 million TPUs to AI corporations Anthropic. If Meta joins as well, TPUs are likely to emerge as a practical alternative to Nvidia's AI chips.
Google, also a major customer of Nvidia chips, has sought to reduce its dependence on Nvidia through in-house TPU development and to expand external supply.
In early this month, Google said it will officially launch the seventh-generation TPU "Ironwood," specialized for inference, within a few weeks. Ironwood delivers up to 4 times the performance of last year's sixth-generation "Trillium" and more than 10 times the performance of the 2023 fifth-generation TPU v5p.
Google TPUs are chips optimized for everything from training large-scale models that require matrix (tensor) operations to complex Reinforcement Learning (RL) and large-scale, low-latency AI inference. Google named the chip a TPU to convey that it is more specialized for matrix operations than general-purpose GPUs or the Neural Processing Unit (NPU).
Google has recently moved aggressively to target the AI market. After unveiling its new AI model "Gemini 3" and challenging ChatGPT's dominance, it also released an image generation and editing tool "Nano Banana Pro" based on Gemini 3.